Vineeth K Kolarath
On 25-10-2026, I spoke to a group of selected students in Kannur about using AI as a mentor, not as an answer machine. I wrote about that idea in Use AI as a Mentor, Not an Answer Machine.
But the part I keep thinking about came after the talk.
Students gathered around with more questions. Not about homework. About the world they are growing up in.
How will AI affect our lives? What does it do to the climate? How can we protect our privacy? And one more: what if AI starts copying itself and builds its own world?
These were good questions. They deserved better than quick answers in a crowded hall.
So I have tried to answer them here, with a little more time to think.
"How will AI affect our lives?"
The honest answer begins with this: nobody knows exactly.
Anyone who tells you precisely what AI will do to jobs, schools or society ten years from now is making a prediction. Some predictions are better informed than others, but none of us can see the future.
What we can say is more useful.
AI is already changing how some tasks are done. It can help write a first draft, translate a paragraph, summarise a long document, organise information or suggest a route. Work that once took people more time can sometimes be completed much faster with these tools.
But a job is rarely just one task.
As I wrote in Automate a Process, Not a Job, a job is a collection of different activities. AI may change some of those activities without replacing the entire job.
A teacher, for example, may use AI to prepare questions or simplify an explanation. But teaching also involves understanding students, recognising when they are struggling and deciding what they need next. Those responsibilities do not disappear just because a machine can prepare a worksheet.
The same is true in many other professions. The work may change, and people will need to learn new ways of doing it.
So my advice to students is simple: do not fear AI, but do not blindly trust it either.
Learn how it works. Understand what it can do well and where it fails. Keep developing your ability to think clearly, solve unfamiliar problems, understand people and take responsibility for decisions.
Do not prepare yourself only to use the tools that exist today. Prepare yourself to keep learning as the tools change.
"Does AI harm the climate?"
This is an important question. It shows that the students were thinking about the costs of technology, not just its convenience.
AI may feel weightless when we use it on a phone or computer. But behind the screen are physical machines.
AI models run on computers, often in large buildings called data centres. Those computers need electricity. They also produce heat, which must be managed. Depending on the cooling system, a data centre may use significant amounts of water as well.
The International Energy Agency (IEA), in its 2025 report Energy and AI, estimated that data centres around the world used about 415 terawatt-hours of electricity in 2024. That was roughly 1.5 per cent of global electricity consumption.
The IEA projected that this figure could rise to around 945 terawatt-hours by 2030 in its main scenario. That would be more than twice the 2024 figure.
One terawatt-hour is a billion kilowatt-hours, or a billion of the electricity units we see on household bills.
There is an important distinction here. These figures are for all data centres, not just those running AI. Data centres also support websites, video services, banking, email and many other digital activities. But the IEA identifies AI as the most important driver of the projected growth in data-centre electricity demand.
So yes, AI has an environmental cost. How large that cost becomes will depend on several things: how much AI is used, how efficient the computers become, how data centres are cooled and, importantly, where their electricity comes from.
There is another side to the story. AI is also being used to improve weather and energy forecasting, help manage electricity grids, integrate renewable energy and make industrial processes more efficient. These applications may help reduce energy use in other parts of the economy.
That does not cancel out the energy AI consumes. Both sides need to be considered.
For a student, the lesson is not to feel guilty every time you ask an AI a question. It is to understand that digital services have a physical cost. Use them thoughtfully, and pay attention to how the energy that powers them is produced.
"How can our privacy be protected?"
This question deserves a very practical answer.
When you type something into an AI service, your message is processed by the systems that operate that service. What happens to it afterwards depends on the provider, the type of account and the privacy settings.
Some services retain conversations for a period of time. Some may use conversations to improve their systems, depending on their policies and your settings. In some circumstances, authorised people may review content for safety, support or service improvement.
So do not assume that everything you type into an AI chat is private.
A few simple habits can help.
Do not share sensitive information.
Avoid entering passwords, bank details, Aadhaar numbers, personal addresses, private family matters or information about other people. Be careful with photographs, school records, examination marks and private documents too.
Understand the settings.
Before using a service regularly, check its privacy controls. Find out whether conversations can be used to improve the service, how long they may be retained and what options you have to delete them or limit their use.
Turning off one setting does not necessarily mean that no information is stored. Read what the service actually says.
Be careful with schoolwork and other people's information.
If a teacher asks you to use an AI tool, it is reasonable to ask what happens to the information you enter. Never upload another student's work, photograph or personal details without appropriate permission.
India has also taken steps to strengthen digital privacy.
The Digital Personal Data Protection Act was passed in 2023, and the Government notified the Digital Personal Data Protection Rules in November 2025. Together, they establish a framework for how organisations handle digital personal data and protect individuals. The framework has a phased implementation timeline, so not all its substantive obligations take effect at once.
Laws and privacy settings matter. But they cannot make every careless decision safe.
The simplest habit is still a good one: think before you share.
"What if AI starts copying itself and builds its own world?"
It sounds like something from a science-fiction film. But researchers do study whether AI systems could copy themselves from one computer to another.
First, let us separate two ideas.
Copying an AI system is not the same as an AI independently deciding to build a world or create a civilisation. Self-replication, in this context, means that an AI-powered system can arrange for its software and the resources it needs to run to be copied or deployed on another computer.
That is a real research question.
In December 2024, researchers from Fudan University in China published a preprint reporting experiments in which two large, openly available language models were able to create working copies of themselves in controlled computer environments.
A preprint is a research paper shared publicly before formal peer review. The experiments involved models being given the tools and conditions needed to attempt the task. They were not evidence that AI systems were independently spreading across the internet.
Since then, researchers have examined the problem more systematically.
For example, the UK's AI Security Institute developed evaluations called RepliBench. These tests examine four capabilities needed for autonomous replication: obtaining computing resources, extracting model weights, deploying a model on another computer and maintaining a persistent presence there.
The evaluations show that AI systems can already complete some of these tasks, although they still struggle to combine them into a reliable process. And success in simplified tests is not the same as success in the real world.
More recently, in October 2026, researchers at Anthropic published experiments involving smaller AI models and inexpensive internet-connected devices. They demonstrated that models could complete important parts of the process in controlled environments, and that some could coordinate the steps needed for replication in simplified tests.
The researchers also made clear that their end-to-end demonstration used a simplified security environment, so it should not be mistaken for proof that the same process would work against properly secured devices.
This is worth paying attention to. But it is important to understand what such experiments do and do not show. They do not establish that AI systems are already spreading independently across ordinary devices or building their own networks in the real world.
So, what should we make of all this?
Today's AI systems do not simply wake up one morning and decide to reproduce themselves. In these experiments, researchers provide goals, tools and controlled environments. However, as systems become more capable of taking actions through software and connected devices, the question of how much freedom they should be given becomes increasingly important.
Who decides what an AI system is allowed to do? What permissions should it have? Who monitors its actions? How can people stop it if something goes wrong?
These are not questions we should dismiss as science fiction. They are engineering, security and policy questions.
The right response is neither panic nor dismissal. It is careful research, sensible safeguards and responsible decisions about how much control we give these systems.
Perhaps one day, some of the students who asked this question will be among the people working on those safeguards.
What these questions told me
I went to Kannur to talk about using AI as a mentor.
The students' questions showed me that they were thinking much further than that.
They were not only asking, "How do I use this?"
They were asking:
What will this do to our lives?
What will it cost the world?
Who can see the information we share?
And how do we remain in control of something so powerful?
Those are important questions. They connect closely with the four questions I discussed in What Is a Thinking Machine?:
What does it notice? How does it decide? What did it learn from? And who checks it?
The students were applying those questions to the world around them.
That is encouraging.
We should not teach young people to be afraid of every new technology. Nor should we encourage them to accept every new invention without question.
We should teach them to be curious, to look for evidence, to recognise uncertainty and to ask who is responsible.
That is how people learn to think about technology rather than simply use it.
Try this
Write down one question about AI that worries you or makes you curious.
Not a homework question. A real one.
Then try to answer it in three parts:
- What do I know for sure?
- What am I still uncertain about?
- What would I need to find out?
You may not arrive at a final answer. That is perfectly fine. Some of the most important questions about AI are still being studied.
In Bring a Rough Idea, I asked you to bring a rough idea.
This time, bring a question.
Sometimes, the best work begins not with an answer, but with a good question.